PerceiverS: A Multi-Scale Perceiver with Effective Segmentation for Long-Term Expressive Symbolic Music Generation

AI-based music generation has made significant progress in recent years. However, generating symbolic music that is both long-structured and expressive remains a significant challenge. In this paper, we propose Perceiver<italic>S</italic> (<italic>S</italic>egmentation and <italic>S</italic>cale), a novel architecture designed to address this issue by leveraging both <bold>Effective</bold> <italic>S</italic><bold>egmentation</bold> and <bold>Multi-</bold><italic>S</italic><bold>cale</bold> attention mechanisms. Our approach enhances symbolic music generation by simultaneously learning long-term structural dependencies and short-term expressive details. By combining cross-attention and self-attention in a <bold>Multi-</bold><italic>S</italic><bold>cale</bold> setting, Perceiver<italic>S</italic> captures long-range musical structure while preserving performance nuances. The proposed model has been evaluated using the Maestro dataset and has demonstrated improvements in generating coherent and diverse music, characterized by both structural consistency and expressive variation.